Unrecognized request argument supplied: seed

[Timmate]

System Info

pandasai v1.4.8, openai v0.27.10, Python 3.10

🐛 Describe the bug

When trying to chat() with a created SmartDataFrame, I'm getting the following error:

'Unfortunately, I was not able to answer your question, because of the following error:\n\nUnrecognized request argument supplied: seed\n'

Apparently, this has to do with v1.4.8 of pandasai and the introduction of the seed param in #742. Obviously, a solution could be simply to downgrade to 1.4.7, but was wondering if there is another way. I'm getting this error for any of the below mentioned Azure OpenAI API versions. @gamcoh, do you have any ideas? Thanks in advance!

Here's the full code:

from pandasai import SmartDataframe
from pandasai.llm import AzureOpenAI

llm = AzureOpenAI(
    deployment_name="gpt35turbotest",
    model_name="gpt-35-turbo",
    api_token=<token>,
    api_base=<azure endpoint>, 
    # api_version="2023-03-15-preview"
    #api_version="2023-07-01-preview"
    api_version="2023-08-01-preview"
)

sdf = SmartDataframe(df, config={"llm": llm})
df.chat(<question>)

[gamcoh]

You're right this is a bug, the openai module should have the latest version in order to use the seed param

[Timmate]

@gamcoh, got it, thanks for your reply! You mean it's a bug in the current release of 1.4.8? Otherwise, I guess I'm missing something, but how is it expected to work now if the openai version specified in poetry.lock is 0.27.10? Tried upgrading openai to the latest version, but this broke pandasai.

[gamcoh]

Yeah same for me, I'm trying to migrate the codebase to openai 1.2.3 but for now I think we need to downgrade

[mspronesti]

#753 introduces support for the newest API (for OpenAI only, no Azure OpenAI yet). Feel free to have a look at it :)

[mspronesti]

The newest version supports OpenAI v1 (including Azure OpenAI) and, among the others, seed. Feel free to have a look :)

[gamcoh]

Great work @mspronesti thanks!

[Timmate]

v1.4.10 (with v1.3.3 of openai) works just fine for me with the above setup (except api_base now is called azure_endpoint) for any of the above Azure OpenAI API versions. Haven't tried setting the seed yet, though -- and, apparently, that's currently impossible with Azure OpenAI, as stated in https://github.com/gventuri/pandas-ai/blob/main/docs/determinism.md.

Closing this as solved anyways.

Many thanks to @mspronesti!